Gait-er-Aid

The path to moving forward.

Comment

Inspiration

The team was inspired to create a solution to the gait (a person's manner of walking) disorder that affects adults and the elderly. Gait-er-aid was created as a product that can be used to analyze the gait of its user using minimal equipment and provide solutions when the gait is suboptimal.

What it does

Gait-er-aid makes use of an RNN (GRU) model that is trained on data obtained from accelerometer, gyroscope, and magnetometer sensors. The RNN model is trained on optimal and sub-optimal gaits and it then uses this to infer when a user uses it to determine whether or not their gait is optimal and suggests a correction to their gait when it's sub-optimal.

How we built it

Gait-er-aid was built with:

  • Accelerometer/Gyroscope/Magnetometer sensors
  • Arduinos
  • Raspberry Pi
  • Qualcomm HDK8450 kit

Programmed Using:

  • PyTorch to define a Gated Recurrent Unit using only 750 seconds of training data to achieve 95% test accuracy and real-time inference.
  • Arduino IDE to read sensor data and write to serial bus.
  • Python data cleaning to filter out faulty data.

Challenges we ran into

The challenges we ran into were:

  • Creating a stable training pipeline to collect data from the sensors. The pipeline to retrieve data from the sensors is: Sensors -> Arduinos -> Raspberry Pi -> Qualcomm HDK8450
  • Calibrating the sensors. We used two slightly different sensors so we had to spend time calibrating both sensors to give similar results when they were being used
  • Communication to the Qualcomm HDK8450 kit
  • Training RNN model with limited training data
  • Labelling the training data with ground truth

Accomplishments that we're proud of

The team is proud of:

  • Creating an RNN model capable of distinguishing an optimal gait from a sub-optimal gait with limited training data.
  • Creating a stable training pipeline

What we learned

The team learned:

  • How to work in a team and utilize the strengths of each team member
  • Capabilities of sequential Neural Networks
  • How to implement real-time inference

What's next for Gait-er-Aid

Better integration into smart devices

Built With

Share this project:

Updates